Apple Blog on GPU Performance of M1Pro and M1Max
mark last edited by mark
I thought some here would be interested in a long and fairly technical blog post about the GPU performance of the new Apple Silicon M1Pro and M1Max.
One thing that I found particularly interesting was this line:
In Photo, an ideal GPU would do three different things well: 1.) High compute performance 2.) Fast on-chip bandwidth 3.) Fast transfer on and off the GPU.
For the machines used by most Isadora users, we've had #1 and #2, but #3 has always been an issue. This third item is crucial when doing things like getting a rendered image from the Stage (i.e., an OpenGL texture that lives on the GPU) back to the the RAM on the CPU. Neither NDI, Blackmagic, nor the system that records the stage can accept an OpenGL texture directly; this means Isadora must pull the texture back to the CPU. (This bottleneck is something I've publicly warned users about repeatedly, and cursed privately because it would be so great to make these this process more efficient.)
The line in the blog post that got me most excited was this, especially the part I've put in bold:
The #M1Max is the fastest GPU we have ever measured in the @affinitybyserif Photo benchmark. It outperforms the W6900X - a $6000, 300W desktop part - because it has immense compute performance, immense on-chip bandwidth and immediate transfer of data on and off the GPU (UMA).
I think the real shift here is that Apple is not looking at a GPU graphics card as a gaming device where data transfer back to the GPU does not need to be fast. Instead, they're looking at the GPU as a video processing device, where transfer to between the GPU and CPU should be as fast as possible in both directions. That shift in perspective could be a game changer.
The "proof is in the pudding" as we say in English, but it's going to be pretty exciting to see how the Apple silicon Native Isadora 3.1.1 performs on these machines.